OpenAI and Poland’s NASK agree to collaborate on AI safety
The partnership will evaluate advanced AI systems and develop safeguards, building on earlier cybersecurity work involving CERT Polska.
OpenAI said it has reached an agreement with Poland’s NASK National Research Institute to jointly examine the safety and security of increasingly capable AI systems. The work will involve testing systems, identifying weaknesses and strengthening protections.
The collaboration will involve NASK’s AI Safety Research Center, which functions as Poland’s AI safety institute. Based in Warsaw and owned by the state, NASK also runs CERT Polska, the team that responds to cyberattacks affecting Polish networks.
The agreement extends an existing relationship. According to NASK, CERT Polska obtained access in July to cybersecurity models from OpenAI through its Daybreak program. OpenAI said those tools helped the team find six flaws affecting software for MikroTik routers. All six have been fixed, the company said.
The deal follows disclosures by OpenAI and Anthropic about unexpected or unwanted behavior by their autonomous agents. In August, OpenAI said some agents under testing had broken out of a controlled environment and jointly attacked Hugging Face, an AI model-sharing platform. That incident has led to a lawsuit against OpenAI in San Francisco.
OpenAI has also urged the United States to coordinate international technical standards for the most advanced AI systems. Its September proposal covered human supervision when AI conducts research and guidance for companies reporting safety incidents.
